An Examination of The Role of Infrastructure
in Credit Lending
Yunkai Ruan
August 19, 2020
Abstract
The positive effect of of infrastructure on credit lending is slow in coming, while the negative effect for districts in large states happens rather soon.
1
Introduction
The casual effect of infrastructure on bank credit lending is seldom studied. The difficulty of determining the casual effect (or lack thereof) that infras-tructure development has on bank credit lending stems from the fact that both development of infrastructure and bank credit lending are an appar-ent outcome of infrastructure borrowing. In order to build infrastructures, government or firms have to borrow money from banks. This infrastructure borrowing will increase bank credit lending. Because the purpose of infras-tructure borrowing is to build infrasinfras-tructures, it should result in a higher infrastructure development. Through their individual relationship with in-frastructure borrowing, inin-frastructure development and bank credit lending could increase in tandem, but this does not indicate a causal effect of infras-tructure development on bank credit lending. It is therefore crucial to find variation in infrastructure development exogenous to infrastructure borrow-ing as well as other omitted variables that could impact credit lendborrow-ing, to be able to consistently estimate the causal effect infrastructure development has on bank credit lending.
common type of infrastructure project in the emerging economy, which cre-ates comparability between districts, as all of them will have observations. Furthermore, instrumental variables will be used to get rid of the impact of endogeneity, so it will be easier to have one type of infrastructure project for the ease of instrumental variable selection.
West Bengal, from the second largest agricultural state in India, experienced rainfall 82% of the normal level in 2012. Punjab, the third largest agri-cultural state in India, had 83% and 53% of the normal level of rainfall in 2011 and 2012 respectively. Because the effect of these large variations in rainfall in large agricultural states should not be ignored, IV regression was run with 12 largest agricultural states dropped. With 12 largest agricultural states dropped, rainfall has a limited impact on agriculture. Without the limited impact on agriculture, and the relatively little impact of rainfall on the other industries, we can be more assured that its impact on credit lending is through road infrastructure.
A district and time fixed effect is run first without the instrumental vari-able. Ideally, district wise credit lending growth is regressed onto district wise growth in road length. However, Aggregate growth in road length in the dis-trict level is unfortunately unavailable. Aggregate growth in road length in the state level is used 1. Districts wise aggregate credit lending growth is regressed onto aggregate road length of state where district resides. The regression is run for up to 4 lags to capture any potential delayed effect in-frastructure has on credit lending. Only the third lag shows a significant positive relationship between infrastructure and credit lending growth. This suggests that infrastructure has a delayed positive impact on credit lending. To ensure the casual effect, the same regression is run again, only this time with logged level of annual rainfall instrumenting for growth in road length. To correct for the potential confounding effect of rainfall influencing credit lending through agriculture, the regression is run with 12 largest agri-culture states dropped. Surprisingly even though the third lag still shows a significantly positive relationship between infrastructure and credit lending, the first lag is now significantly negative , which means an increased growth
1district is a geographical area one level smaller than states. One state comprises of
of state aggregate road length decreases the district aggregate growth rate in the short term. Because our growth in road length is aggregated at the state level, it is well probable that this increase in growth concentrated in a few regions due to the nature of road construction that connects strategic locations and important cities (Ghani, Goswami, & Kerr, 2017). A decrease in district wise credit lending growth reflects that on average districts see their credit lending growth slows down. This could be the case when most of the districts are not close the new roads and therefore hurt by the negative spillover effect of infrastructure: An increase of infrastructure benefits ares surrounding the infrastructure but harms the further ares in the short term since investors and entrepreneurs move closer to the infrastructure at the expense of further areas. (Holl, 2004) (Boarnet, 1998).
To verify this negative spillover effect, states that are not the top 12 states in agricultural production are divided into large states and small states. Large states have 25.4 districts each on average while small states on average have 4.6 districts on average. Since increase in state wise aggregate road length will likely to concentrate in a few districts, most districts in large states will be hurt by the negative spillover effect. For small states with only few districts, an increase in state wise aggregate road length will likely to benefit most of their districts. In some cases, union territories (also counted as a state) have only one district and an increase in state wise aggregate road length will undoubtedly be in this district and should not cause negative spill over effect captured by district wise credit lending. The results are convincing: Growth of road length in large states impacts district credit lending growth negatively the next year while growth of road length in small states does not render such a negative impact. Moreever, districts from both large and small states see their credit lending grows in response to road development in 3 and 4 years.
on credit lending in all cases is slow in coming, while a negative impact happens immediately in districts from large states, presumedly due to the negative spill over effect.
The rest of the paper will be laid out as the following: Section 2 will talks about the literatures and phenomenon that motivate the study of link between infrastructure development and credit lending as well as the hypoth-esis development. Section 3 will look at the data essential for the analysis. Section 4 will provide the methodology by which the analysis is run. Section 5 will interpret these results, what they mean for policy makers, limitation in the research as well as ideas for further exploration in the topic. Finally, Section 6 will be the conclusion.
2
Motivation and hypothesis
2.1
Empirical evidence of positive impact of credit
lend-ing on GDP growth
Theoretically, more lending, under the right conditions, enables more invest-ment activities which could lead to higher GDP growth. In their paper Im-pact of World Bank lending in an Adjustment-Led Growth Model, Sushanta Mallick and Tomoe Moore sampled 30 countries who are the recipients of loans from the World Bank, and found that in the long run, lending helps boost economic expansion conditional on macroeconomic and political envi-ronment (Mallick & Moore, 2005).
2.2
Empirical evidence of positive impact of
infras-tructure development on GDP growth
Studies over the effect of infrastructure on economic growth have been done all over the world. In the long run, better infrastructure leads to higher level of output (Canning & Fay, 1993) (Mbulawa, 2017).The the positive effect of infrastructure is quite robust and universal: a study of 96 countries from 1960-1985 found cross sectional evidence of higher level of GDP due to the higher amount of transportation infrastructure (Canning & Fay, 1993). Infrastructure used can be diverse and unconventional: Telephones and power plants are used to measure the level of infrastructure in 75 countries over the period of three decades found that the contribution of these infrastructure is more than its cost to the GDP(Esfahani & Ramirez, 2003).
2.3
Empirical evidence of negative spill over effect of
road infrastructure
While there are strong evidence that infrastructure has a positive impact in output. Negative impact of infrastructure at the form of spill over effect is recorded as well. Increase in infrastructure capital increases output in the county where the development occurs but decreases output in other coun-ties (Boarnet, 1998). Firms prefer locations close to new motorways at the expense of further locations (Holl, 2004) .
2.4
Hypothesis
Because infrastructure development in the long run leads to higher output, which is associated with a higher level of credit lending, I hypothesize that infrastructure development increases credit lending.
Hypothesis : Infrastructure development increases credit lend-ing.
A negative spillover effect will happen in the most districts in big states because big states have many districts, most of which are likely not in the strategic infrastructure development that connects important cities. Firms are attracted by new infrastructures and reduce investing in districts where new infrastructures are not located. Since most districts suffer from negative spillover while only a small number of districts benefit, the average effect of state road development on district credit lending will be negative.
3
Data
This research involves a panel data from 700+ districts in India ranging from year 2007 to 2015. Data for credit lending are found in the Reserve Bank of India data base. Road length and rainfall statistics are found in the Ministry of Statistics and Programme Implementation in India. Credit outstanding level, our dependent variable of interest, are inflation adjusted to the 2007 level.
Variables of interest will be listed and explained.
3.1
District wise aggregate credit lending
District wise aggregate credit level is the sum of credit outstanding in the asset part of balance sheet from all banks in a district. District is a subset of state. All banks include private banks, public banks, regional rural banks and foreign banks.
3.2
Growth in credit lending
Growth in credit lending is calculated using the aggregate credit lending data :
gcit =
cit− cit−1
cit−1
(1) where gcit is the annual growth rate of aggregate credit at time t in district
i, and citis the district wise aggregate credit outstanding level at time t in
district i.
3.3
State wise aggregate road length
many districts while some union territories only have one.
I aggregated the length of roads by summing up road length from different road types – national highways, state highways, other PWD roads, rural roads and project roads – within a state. Data for all road types, except for rural roads, are available from 2007-2014. Data for rural roads, which make up around 70% of total road length are only available from 2011-2014. Therefore, data for state wise aggregate road length range from 2011-2014.
3.4
Growth in road length
Growth in road length is calculated using the aggregate road length data, similar to the calculation of growth in credit lending:
grit =
rit− rit−1
rlit−1
(2) where grit is the growth in length of the roads at time t in state i, rit is
aggregate state road length at time t in statei.
3.5
Annual rainfall – an instrument for infrastructure
development
Table 1: Summary of variables
This table reports the summary statistics of growth in credit at the district-year level, while other data are at the state-district-year level. Growth in credit and annual rainfall observations cover more years to allow regressions done with lags.
(1) (2) (3) (4) (5) (6)
Variables N mean sd min p50 max
Annual growth in credit 6,932 0.0776 0.0907 -0.209 0.0761 0.400 Annual rainfall level (log) 5,877 6.961 0.493 5.722 6.902 8.274 Annual growth in total roads 1,950 0.0647 0.150 -0.319 0.0323 1.668 Annual growth in paved roads 1,950 0.118 0.416 -0.340 0.0447 3.175 Annual growth in unpaved roads 1,934 0.0797 0.436 -0.878 0.00155 3.214
for the year 2007. Uttar Pradesh has a half of its area in the Uttar Pradesh East meteorological division and a half of its area in the Uttar Pradesh west meteorological division. The annual rainfall for the state is the average of 586.1 millimeters( from Uttar Pradesh West) and 863.2 millimeters (from Uttar Pradesh East).
3.6
Summary Statistics
4
Methodology
Since it is a panel data, growth of credit in the same district through time are likely to be influenced by district specific time invariant factors such as work culture and average age of the work force, which at the same time might be correlated with road development. . This will create the problem of endogeneity and bias the results. On the other hand, growth of credit in the same year throughout districts are likely to be influenced by time specific factors such as broader economic climate, which could also affect credit lending. This could confound the effect of road infrastructure on credit lending. Therefore, district and year fixed effect is used to give each district and time their own constant value in response to these district and time invariant factors.
Annual growth rate of district wise aggregate credit outstanding is re-gressed onto the annual growth rate of aggregate road length of the state where the district resides. This is done up to four lags to try to capture any short term effect growth in roads will have on growth in credit lending:
gcit = ai+ yt+ b ∗ grit−s+ eit, s = 1, 2, 3, 4 (3) and gcit = cit− cit−1 cit−1 , grit= rit− rit−1 rit−1 , (4)
where gcit is the annual growth rate of credit at time t in district i, ai
is the district dependent unobserved heterogeneity, yt is the time dependent
unobserved heterogeneity, grit−s is the growth in length of the road at time
t − s in state i , citis the district wise bank aggregate credit outstanding level
at time t and rit is aggregate state road length at time t.
could cofound the true effect of road infrastructure on credit outstanding. One obvious example of such a correlation will be the effect of credit lending on road infrastructure (reversed causality). If credit lending leads to longer road because banks give out loans for road construction, it will create the problem of endogeneity. Instrumental variable is used to to create the pre-dicted value of state wise road length growth which does not covary with any factors that could have an effect on the district wise aggregate credit outstanding.
Annual rainfall level is used as an instrumental variable to tackle the above mentioned difficulties due the these desirable effects : 1) Annual rain fall affects the growth of road length significantly 2) Annual rainfall level is exogenous by nature and will therefore render a good exogenous source of variation in growth in road length.
4.1
Empirical evidence for annual rainfall as an
instru-mental variable
Aziz and Abdel-Hakam found 290 causes for delay of road construction after literature review and of which the number 1 most frequently mentioned rea-son is weather condition (Aziz & Abdel-Hakam, 2016). Of the top 20 causes, weather condition is one of the two causes which are categorized as exter-nal 2, which should be the ideal candidates for an exogenous instrumental variable. Major weather conditions that impact road construction are snow and rain(Lang, 1978). Since rain happens much more broadly in India than snow, rainfall is used as an instrumental variable.
Many scientific researches have been done on the effect of rainfall on road construction. Construction work is one of the most heavily impacted industry by seasonal variation and highway construction consists of a substantial part
of this industry(Lang, 1978) . Due to its significant impact on agriculture, rainfall ,when used as an instrumental variable, requires the iv regressions to run with 12 largest agricultural states dropped to limit the its impact on credit lending through agriculture. Dropping these agricultural states makes rainfall a viable instrumental variable as it will mostly impact the credit lending through our road growth.
The research board on transportation also identified the main problem facing road construction under the rain to be the increased moisture content
in the pavement material that will affect the compaction process. Thus
during rainy weather, construction of paved roads has to be halted. To
obtained the empirical evidence of said effect rainfall has on road pavement, I regressed the state wise growth in roads (paved, unpaved and total roads) length separately onto the lagged 1 log level of state wise annual rainfall. Because rainfall might be associated with geographical elevation – a time invariant variable which could affect road development – state fixed effect is used. Moreover, rainfall could be associated with global warming which might impact governmental policy on road development, time fixed effect is also used in conjunction with state fixed effect.
grit= ai+ ct+ b ∗ log(rfit−1) + eit (5) and grit = rit− rit−1 rit−1 (6) where grit is the growth in length of the roads, ai is the state dependent
unobserved heterogeneity, ctis the time dependent unobserved heterogeneity,
rfit−1 is the annual rainfall level in the previous time period, and ritis the
state wise length of road infrastructure.
The regression results are shown in table 2:
Table 2: Impact of annual rainfall on the development of different types of roads
This table reports results of the impact of annual rainfall on growth of ag-gregate, paved and unpaved road length respectively. Data for road growth are at the state level and range from 2012-2014. Data for annual rainfall are at the state level as well and are logged and lagged by one year.
(1) (2) (3)
Variables Growth rate of aggregatel roads paved roads unpaved roads
Log (annual rainfall) -0.197*** -0.687*** -0.041
(0.058) (0.090) (0.093)
Constant 1.442*** 4.918*** 0.366
(0.404) (0.626) (0.653)
Observations 1,928 1,928 1,911
R-squared 0.380 0.412 0.348
statistically significant impact on the growth of unpaved roads as indicated by column (3). This result corresponds well with the road infrastructure research done by the U.S. transportation research board that the main delay rainfall causes on road construction involves the increased moisture level of the pavement material caused by rain. Since unpaved roads do not use pavement material, the construction of which should not face any major problems under rain. Overall, due to its impact on paved roads, rainfall impacts the growth rate of aggregate road length. Lastly, annual rainfall is exogenous by nature and impacts linearly road construction mainly. It should explain the exogenous variation in road length and annual growth in road length.
5
Results
5.1
Road development increases credit lending with
delay
Table 3: Impact of growth in road length on credit lending growth, reduced form
This table reports the results of the impact growth in roads has on credit lending for the next four years after road development is completed. Data for road growth are at the state level and range from 2012-2014. Data for credit lending are at the district level and range from 2013-2018 due to lagged regressions.
(1) (2) (3) (4) Variables Credit growth rate in 1 year In 2 years In 3 years In 4 years Growth rate of roads -0.021 -0.017 0.033** 0.035**
(0.018) (0.019) (0.016) (0.016) Constant 0.076*** 0.089*** 0.087***
(0.001) (0.001) (0.001) Observations 1,906 1,915 1,924 1,919 R-squared 0.486 0.001 0.505 0.486
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
on top of 6.5% will increase the growth rate by 16.4 times3. A 16.4 times
increase of road building speed will results in a 1.88 times increase in growth of credit lending at the current situation4. The economic significance of road construction in terms of promoting credit lending is small, at least for the first four years after the road construction is completed.
To correct for endogeneity, the same regressions are run but this time with 12 largest agricultural states dropped and independent variables in-strumented by logged level of annual rainfall. Reducing endogenity found negative spillover effectt: As shown in table 4, across districts and time, an
3(100%+6.5%)/6.5%=16.4
increase in the growth rate in the length of total roads has a negative effect on the growth of credit lending immediately the next year and a positive im-pact of growth in road length on the growth in credit in year 3. The negative impact happens almost immediately and carries a bigger magnitude than the positive impact which happens later. An addition of 100% in growth rate of road length subtracts 16% off the growth rate of credit lending the next year. Despite an increase in credit growth rate at the third year, the overall impact after 4 years remain negative. Although the increase in credit growth rate at the third year is consistent with the delayed impact of road development on credit lending, the immediately negative impact at the first lag is indicating a negative spillover effect. This is especially possible given our independent variable is at the state level. An increase of road growth at the state level is likely connecting strategic locations in a few big and important districts, the construction of new roads in these few districts halts investment in the other districts because investors are attracted by new roads. They plan to invest in districts where new roads are built at the expense of districts without these new roads. Even though the positive delayed effect kicks in later as invest-ments plans are realized, this comes much later and weaker in magnitude as compared to the sudden halting of investment.
5.2
A double check for the negative spillover effect
Table 4: Impact of growth in road length on credit lending growth, IV estimation
This table reports the results of the impact growth in roads has on credit lending for the next four years after road development is completed, with growth in roads estimated by lagged 1 logged level of annual rainfall. Data for road growth are at the state level and range from 2012-2014. Data for credit lending are at the district level and range from 2013-2018 due to lagged regressions. Data for rainfall level are at the state level and range from 2011-2013.
(1) (2) (3) (4) Variables Credit growth rate in 1 year In 2 years In 3 years In 4 years Growth rate in roads -0.160** 0.008 0.103* -0.037
(0.063) (0.050) (0.061) (0.072) Observations 932 931 933 926 IV estimation Yes Yes Yes Yes
4.6 districts on average. Results are shown in table 5 column 1 : Big states are impacted by the negative spillover effect strongly while the statistically insignificant coefficient with much smaller magnitude indicates that small states are not impacted by the same effect as much.
5.3
Why does the negative spillover effect happen
al-most immediately while the positive impact
hap-pens with a delay
it is easier to stop doing something than to start doing something. When new roads are built to connect strategically important districts, investors start planning to take advantage of this new development as they would like to be close to areas with new roads. During planning, it will not make sense to keep investing in districts where new roads are not located, this investment immediately stops which causes a decrease in credit lending growth. Stop-ping investing is relatively easy to do. On the other hand, new investment plans could take years to formulate, after which investors might go through a lengthy period of credit approval process from banks. Only after all of these are done will credit lending growth be affected. This explains the delayed positive impact of road construction on credit lending.
5.4
Limitation
Table 5: Impact of growth in road length on credit lending growth in big states
This table reports the results of the impact growth in roads has on credit lending in big and small states separately. Big state group has each state with 25,4 districts on average, while small state group has each state with 4.6 districts on average. Growth in roads is estimated by lagged 1 logged level of annual rainfall. Data for road growth are at the state level and range from 2012-2014. Data for credit lending are at the district level and range from 2013-2018 due to lagged regressions. Data for rainfall level are at the state level and range from 2011-2013.
Big states
(1) (2) (3) (4) Variables Credit growth rate in 1 year In 2 years In 3 years In 4 years Growth rate in roads -0.237*** -0.002 0.220*** -0.115*
(0.063) (0.044) (0.060) (0.059) Observations 762 761 757 750 IV estimation Yes Yes Yes Yes
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Small states
(1) (2) (3) (4) Variables Credit growth rate in 1 year In 2 years In 3 years In 4 years Growth rate in roads -0.049 0.112 -0.177 0.150***
(0.086) (0.135) (0.119) (0.047) Observations 170 170 176 176 IV estimation Yes Yes Yes Yes
Road length is used as a proxy for infrastructure. Although it is reason-able to assume other infrastructures could impact credit lending similarly, this assumption is not 100% guaranteed to be valid.
6
Conclusion
State Groups
Out of non farm states: Out of non farm states: All Sates Non farm states Large states Small states Andaman & Nicobar Islands Andaman & Nicobar Islands Arunachal pradesh Andaman & Nicobar Islands
Andhra pradesh Arunachal pradesh Bihar Chandigarh Arunachal pradesh Bihar Himachal pradesh Dadra & nagar haveli
Assam Chandigarh Jammu & kashmir Daman & diu Bihar Dadra & nagar haveli Jharkhand Goa Chandigarh Daman & diu Maharashtra Lakshadweep
Chhattisgarh Goa Odisha Manipur
Dadra & nagar haveli Himachal pradesh Rajasthan Meghalaya Daman & diu Jammu & kashmir Tamil nadu Mizoram
Goa Jharkhand Uttarakhand Nagaland
Gujarat Lakshadweep Nct of delhi
Haryana Maharashtra Puducherry
Himachal pradesh Manipur Sikkim
Jammu & kashmir Meghalaya Tripura
Jharkhand Mizoram
Karnataka Nagaland
Kerala Nct of delhi
Lakshadweep Odisha
Madhya pradesh Puducherry Maharashtra Rajasthan
Manipur Sikkim
Meghalaya Tamil nadu
References
Auffhammer, M., Ramanathan, V., & Vincent, J. R. (2012). Climate change, the monsoon, and rice yield in india. Climatic change, 111 (2), 411– 424.
Aziz, R. F., & Abdel-Hakam, A. A. (2016). Exploring delay causes of road construction projects in egypt. Alexandria Engineering Journal , 55 (2), 1515–1539.
Boarnet, M. G. (1998). Spillovers and the locational effects of public infras-tructure. Journal of regional science, 38 (3), 381–400.
Canning, D., & Fay, M. (1993). The effects of transportation networks on economic growth.
Esfahani, H. S., & Ramirez, M. T. (2003). Institutions, infrastructure, and economic growth. Journal of development Economics, 70 (2), 443–477. Gadgil, S., & Gadgil, S. (2006). The indian monsoon, gdp and agriculture.
Economic and Political Weekly, 4887–4895.
Ghani, E., Goswami, A. G., & Kerr, W. R. (2017). Highways and spatial location within cities: Evidence from india. The World Bank Economic Review , 30 (Supplement 1), S97–S108.
Holl, A. (2004). Manufacturing location and impacts of road transport
infrastructure: empirical evidence from spain. Regional Science and Urban Economics, 34 (3), 341–363.
Jodha, N. (1981). Role of credit in farmers’ adjustment against risk in arid and semi-arid tropical areas of india. Economic and Political Weekly, 1696–1709.
Lang, A. S. (1978). Effect of weather on highway construction.
Mallick, S., & Moore, T. (2005). Impact of world bank lending in an
economic growth in botswana.
